CacheTiltFABLE 5 → 5.1
AN INDEPENDENT COST STUDY

Fable 5 vs 5.1.
What changes your cost?

Start with a synthetic example. Choose a category to change its tokens. Separate the new rate from a different amount of work.

Fable 5 baseline spend: $42.50. Cache reads account for $20 of that spend.
Lower token cost35.3%

$42.50at Fable 5 rates, compared with$27.50at Fable 5.1 rates

USD · same counts

The ring divides Fable 5 baseline dollars, not tokens. The text beside each category gives its subtotal rounded to cents.

One change. Two causes.

Rate effect−$15.00Original usage, new rates

Usage effect$0.00Changed usage, new rates

Net change−$15.00New cost minus old cost

The cache-read rate changes $42.50 to $27.50. Counts are unchanged, so the usage effect is zero.

Change the example, not the story.

Each preset is invented to isolate one cause. None is a measured model benchmark.

For the starting counts, 300,000 additional output tokens would exactly use up the $15 cache saving, if every other count stayed fixed.

Why the 75% discount becomes 35.3% here

See all token counts and category costs
Original synthetic workload · all amounts in USD
Category5 tokens5 cost5.1 tokens5.1 cost
Fresh input1,000,000$10.001,000,000$10.00
5-minute writes200,000$2.50200,000$2.50
1-hour writes0$0.000$0.00
Cache reads20,000,000$20.0020,000,000$5.00
Output200,000$10.00200,000$10.00

Displayed money is rounded to cents; the download retains exact totals. Tiny changes can be below one cent.

Count a comparable piece of work.

Use two completed runs of the same task population if you want to compare real usage. A week with twice the traffic is a different comparison from the same requests on a new model. Include every request and retry in the chosen scope; do not mix averages per call with totals per task.

This instrument prices only the five token categories shown. It does not predict how either model will behave, connect to an account, explain subscription limits or calculate a complete invoice. Standard global Claude API list rates, checked . Counter mapping, rates and exclusions.